Speculative Design for an AI Noise Machine: Signal, Noise, and the Aesthetics of Algorithmic Ambience

Presenter: Paulus van Horne (Boulder, US)

Abstract

This presentation examines the productive tension of applying generative AI – a technology fundamentally oriented toward maximising signal-to-noise ratio – to the creation of ambient noise. I detail a speculative plan for an AI-powered noise machine that interrogates the boundary between signal and noise in both sonic and computational contexts.  The work positions the AI noise machine within frameworks of functional sound design while questioning our relationships to algorithmic medi(t)ation and the politics of auditory attention in domestic spaces. Drawing from media archaeology of popular noise machines like the Marpac Sleepmate, dOhm and Lectrofan, I discuss how these devices achieve effectiveness through carefully crafted non-repetition and recognisable yet non-specific sonic signatures. Generative AI complicates this paradigm, operating in the liminal space between coherent output and computational hallucination.